Raschka's final Reasoning From Scratch lecture covers distilling reasoning models

Sebastian Raschka posts the eighth and final lecture of his Reasoning From Scratch series on X, covering distillation as a more efficient alternative to reinforcement learning training runs.

The session covers soft versus hard distillation and on-policy versus off-policy approaches, and walks through training Qwen3-0.6B on DeepSeek R1 MATH reasoning traces with evaluation on MATH-500. Raschka says he plans a brief break after the series and will be back.